US 20060004765 A1
Addressed is a system and method for remote data caching and replication by local copy maintenance of remote data within a SAN file system. Distributed Storage Tank (DST), an extension to a SAN file system, provides for transparent SAN client access of local copies by importing, exporting, and storing data using network file access protocols as well as by providing assurance of metadata and file content validity. A Remote Access Agent (RAA) handles protocol implementation and conversion necessary for communication with remote data sources. Controlled by a consistency policy, consistency is maintained by RAA fetching and updating local copies if modifications have occurred to a file since it was first stored as a local copy in local storage. Additionally, RAA returns metadata pertaining to the requested data. A SAN client obtains metadata corresponding to the requested data and utilizes it to directly access locally stored copies of remote data.
1. A method for storing and accessing remotely obtained data as a copy local to a Storage Area Network (SAN), wherein a Remote Access Agent (RAA):
a. obtains remote data from a remote source; said remote source external to said SAN;
b. decouples attributes from content in said obtained remote data;
c. locally stores said decoupled attributes and a copy of said decoupled content, separately; and
d. utilizes said locally stored attributes to obtain file-level access to said local copy.
2. A method for storing and accessing, as per
3. A method for storing and accessing, as per
4. A method for storing and accessing, as per
5. A method for storing and accessing, as per
6. A method for storing and accessing, as per
7. A method for storing and accessing, as per
8. A method for storing and accessing, as per
a. obtaining said metadata associated with said local copy;
b. extracting said content storage attribute from said obtained metadata; and
c. reading from or writing to a physical storage location in said SAN disk;
said physical location specified by said content storage attribute.
9. A method for maintaining consistency, in a Storage Area Network (SAN), of a local copy of a remote file system sub-tree obtained from a remote source; said method comprising:
a. mapping directory structure of said remote file system sub-tree to a remote container; said remote container attached to said SAN;
b. representing each remote object comprising said remote file system sub-tree as a local object component of said remote container;
c. labeling each of said local objects with attributes associated with said represented remote object;
d. storing metadata describing each of said local objects in a metadata server;
e. associating a consistency policy with each of said local objects in said remote container; said policy defining conditions for checking freshness of said labeled attributes; and
f. updating said local object components of remote container in accordance with said consistency policy.
10. A method for maintaining consistency, in a Storage Area Network (SAN), as per
11. A method for maintaining consistency, in a Storage Area Network (SAN), as per
12. A method for maintaining consistency, in a Storage Area Network (SAN), as per
13. A method for maintaining consistency, in a Storage Area Network (SAN), as per
14. A method for maintaining consistency, in a Storage Area Network (SAN), as per
a. storing said file content at a physical location specified by said content storage attribute; if said local object represents a file; and
b. populating said local object with said directory entries; if said local object represents a directory.
15. A method for maintaining consistency, in a Storage Area Network (SAN), as per
a. triggering a check of said labeled attributes associated with said local object;
b. checking for the existence of said metadata defining a consistency guarantee for said local object;
c. fetching from said remote source, attributes associated with said remote object represented by said local object, based on results of said checking step;
d. creating a consistency guarantee for said local object, based on results of said fetching step; and
e. merging said fetched attributes with said attributes associated with said local object.
16. A method for maintaining consistency, in a Storage Area Network (SAN), as per
17. A method for maintaining consistency, in a Storage Area Network (SAN), as per
18. A method for maintaining consistency, in a Storage Area Network (SAN), as per
19. A method for maintaining consistency, in a Storage Area Network (SAN), as per
20. A method for maintaining consistency in a Storage Area Network (SAN), as per
21. A method for maintaining consistency in a Storage Area Network (SAN), as per
22. A method for maintaining consistency in a Storage Area Network (SAN), as per
23. A method for refreshing an object local to a Storage Area Network (SAN) in accordance with a consistency guarantee; said local object representing a remote file system sub-tree; said local object refreshing with a remote object imported from a remote source; said method comprising:
a. retrieving remote data from said remote source;
b. decoupling content and attributes from said remote data;
c. modifying said local object on the basis of said decoupled attributes and content;
d. labeling said local object with said decoupled attributes; and
e. updating metadata associated with said local object.
24. A method for refreshing an object local to a Storage Area Network (SAN), as per
25. A method for updating a remote object at a remote source with modifications made to an object local to a Storage Area Network (SAN); said local object representing a remote file system sub-tree; said update in accordance with an update consistency policy; said method comprising:
a. detecting modifications to a local object; said local object marked as modified;
b. obtaining attributes and content associated with said marked local object;
c. exporting said obtained attributes and content to said remote source; and
d. receiving remote object attributes and a consistency guarantee from said remote source.
26. A method for refreshing an object local to a Storage Area Network (SAN), as per
27. A method for refreshing an object local to a Storage Area Network (SAN), as per
28. A method for refreshing an object local to a Storage Area Network (SAN), as per
29. A system for importing, storing, and exporting, in a Storage Area Network (SAN), a local copy of remote file system data; said system comprising:
a. a client on said SAN requesting, storing, and accessing said local copy;
b. a metadata server, local to said SAN, storing metadata describing attributes of said local copy;
c. a source of remote file system data external to said SAN;
d. a Remote Access Agent (RAA) sending said local copy to and receiving said remote file from, said remote source; a storage device, attached to said SAN, storing content of said local copy.
30. A system for importing, storing, and exporting, as per
31. A system for importing, storing, and exporting, as per
32. A system for importing, storing, and exporting, as per
33. A system for importing, storing, and exporting, as per
34. A system for importing, storing, and exporting, as per
35. A system for importing, storing, and exporting, as per
36. A system for importing, storing, and exporting, as per
37. A system for importing, storing, and exporting, as per
38. A system for importing, storing, and exporting, as per
39. A system for importing, storing, and exporting, as per
40. A system for importing, storing, and exporting, as per
41. A system for updating, storing, and accessing, in a Storage Area Network (SAN), a local copy of file system data from a remote source; said system comprising:
a. a client interface, connected to a SAN client, sending to and receiving from said SAN client, metadata and a request for access to said local copy;
b. a remote container module connected to said client interface; said remote container module managing consistency of said local copy and processing said request;
c. a cache manager connected to said remote container module managing and allocating storage space for said local copy;
d. a replica manager connected to said remote container module; said replica manager maintaining consistency of said local copy if validation is asynchronous type;
e. a distributed file manager connected to said remote container module; said distributed file manager updating said local copy by obtaining said remote source file system data; and
f. a consistency policy maintenance framework connected to: said remote container module, said distributed file manager; said replica manager; said cache manager storing conditions governing consistency management of said local copy.
42. A system for updating, storing, and accessing, in a Storage Area Network (SAN), as per
43. A system for updating, storing, and accessing, in a Storage Area Network (SAN), as per
44. A system for updating, storing, and accessing, in a Storage Area Network (SAN), as per
45. A system for updating, storing, and accessing, in a Storage Area Network (SAN), as per
46. A system for updating, storing, and accessing, in a Storage Area Network (SAN), as per
47. A system for updating, storing, and accessing, in a Storage Area Network (SAN), as per
48. A system for updating, storing, and accessing, in a Storage Area Network (SAN), as per
49. A system for updating, storing, and accessing, in a Storage Area Network (SAN), as per
50. A system for updating, storing, and accessing, in a Storage Area Network (SAN), as per
51. A system for updating, storing, and accessing, in a Storage Area Network (SAN), as per
1. Field of Invention
The present invention relates generally to the field of file sharing. More specifically, the present invention is related to remote file system caching.
2. Discussion of Prior Art
In a Storage Area Network (SAN), a SAN file system is used to provide high-performance access to large file systems. A SAN file system is a type of cluster file system in which both clients and servers are connected to a SAN. Sometimes client and server roles are combined symmetrically in a single machine, as in the case of GPFS, and sometimes these functional roles are partitioned onto distinct machines, as in the case of Storage Tank (ST). Two key characteristics of SANs affect systems in which all participants are required to have SAN access. First, a SAN has a restricted geographical extent, thus limiting it to a relatively small campus. Second, participating hosts must be trusted not to corrupt file system data.
However, users and applications are often distributed over a large area or are contained within administrative domains that do not fully trust each other. In these situations, a bridge is needed to allow file sharing across Wide Area Networks (WAN), administrative domains, or both. The two ends of the bridge, namely, the import and export sides, generally have different characteristics. The bridge can be constructed at the file-level above a SAN file system or at the block-level below it. Issues requiring consideration include security, performance, and administration. A WAN admits a much larger range of trust and has different threat models than a SAN. Therefore, strong security and a flexible approach to integration and interface with multiple domains are necessary to ensure data availability and authenticity. Additional provisions are necessary to account for performance parameters unique to a WAN, in specific, higher latencies and limited bandwidth.
Issues associated with exporting data from a SAN file system can be addressed by using an existing distributed file system protocol. For example, a node of a General Parallel File System (GPFS) could run a Network File System (NFS) or Samba server, which could, in turn, deliver SAN file system data to remote users. These protocols are widely implemented, but are not well adapted to a WAN environment. A more suitable protocol, such as Andrew File System (AFS) or Distributed File System (DFS), could be used. While a DFS server has been integrated, albeit in a limited way, with GPFS on AIX, it is not available for any other SAN file systems, nor has AFS been ported to such an environment.
A complementary task is importing remote data into a SAN file system so that local applications have access to it. Machines acting as clients of a SAN file system (e.g., ST clients or GPFS nodes) can individually run distributed file system clients, such as NFS or AFS, but forgo many of the benefits of a SAN file system. Features missing from these discrete solutions include centralized administration and management, uniform namespace, shared caching, and reduced WAN usage.
Another approach to circumventing SAN distance limitations is to use an IP-compatible storage protocol such as Internet Small Computer System Interface (iSCSI). This gives a SAN file system the capability to expand geographically and include constituents distributed over a WAN. Issues associated with this approach are generally due to performance and scaling. Software for storage systems designed to benefit from low latencies of a SAN may require redesign when faced with much larger WAN latencies and an expanded set of more complex failure modes. A SAN file system that spans a large geographical area will generally have more network constituents, which can challenge the scaling abilities of a file system's architecture. Specifically, cluster algorithms controlling file system metadata consistency may behave poorly when the number of network constituents increases. Additionally, distributed network constituents may necessitate the use of security model that is not as restricted as that of a SAN.
Therefore, there is a need in the art to address data importation from a remote source and data exportation to a remote source. There are at least two problems with performing these tasks within the context of a file server, specifically, within the same process. First, transferring file data imposes large network and disk loads on the server, which can severely impact its performance on other tasks. Second, supporting multiple protocols necessary to communicate with a variety of remote sources can lead to software compatibility issues, maintenance problems, and large executable footprints.
Existing strategies that address some aspects of this concern by decoupling metadata handling from the burden of transferring file contents include File Transfer Protocol (FTP), SAN third-party backup, and HSM using DMAPI. FTP transfers data using distinct control and data channels. Thus, a separate agent operating at the file-level handles file contents. However, because FTP operates at the file-level, it is limited in that it cannot support decoupled metadata and file contents in a SAN File System.
SAN-based backup systems, such as Veritas™, Pathlight™, and Legato/Celestra™, alleviate server load by providing server-less backup. These systems utilize a SAN environment to copy data between storage devices, such as disk to tape, using a block-level interface such as Small Computer System Interface (SCSI). They are limited, however, in that they do not provide for file system interaction and thus, do not provide file-level access to data residing on either a SAN or across a WAN.
Hierarchical storage management systems gained additional portability and interchangeability with the development of the Data Management API (DMAPI). This interface allows the kernel to intercept references to selected portions of a file system and pass them to a user space DM application. Access to data is allowed at the file-level through the file system, so it does not support a remote DM application nor does it take advantage of a SAN environment to offload file server machines.
The present invention has wide applicability since it makes limited assumptions about the architecture of a SAN file system and does not make extreme performance demands on a WAN.
Whatever the precise merits, features, and advantages of the above cited references, none of them achieves or fulfills the purposes of the present invention.
The system and method of the present invention address remote data import and export by integrating caching and replication functions within a SAN file system and by maintaining local copies of remote content on local SAN disks. Remote data, which is comprised of foreign data from file systems external to a SAN and distant data available over a wide-area network (WAN), is supplied to SAN clients via a Distributed Storage Tank (DST). Thus, a SAN file system acts as a caching proxy. Frequent SAN client access of remote data is facilitated by the sharing of local copies of remote data on SAN disks. To obtain access to shared local copies, a DST first fetches remote data using network file access protocols necessary for communication with a remote source. Remote data is then stored and maintained on a SAN disk, as a local copy internal to a SAN. Revalidation and refresh operations occur on local copies of remote data in accordance with a consistency policy. A consistency policy database is consulted to determine a validation consistency level for the maintenance of local copies of remote data, where validation consistency is defined as the frequency of validation operations, for example, checking file attributes to determine if a local copy is the same as its corresponding remote copy. Validation operations are performed to determine if a local copy is the same as a remote copy. Validation consistency levels of specified granularity are chosen by a system or user policy and stored in a policy framework. In a preferred embodiment, a server agrees to inform a client when changes affect a resident local copy of remote data.
DST is an extension to a SAN file system that provides access to remote file systems. DST manages file system objects by caching and replicating data from remote sources as local copies in SAN disks and by allowing shared, SAN client access to local copies. DST fetches remote file system objects using multiple network file access protocols. A SAN metadata server (MDS) provides an interface for accessing metadata and file contents of remotely obtained objects. Additionally, a provision is made for the specification of consistency guarantees, which provide assurance of metadata and file content validity. In this manner, consistency policy requirements and network file access protocol limitations are balanced. DST is comprised of several modules; a remote container module, a distributed file manager, a plurality of remote access agents, a cache manager, a replica manager, and a consistency maintenance framework. With the exception of a remote access agent, these components run on SAN server machines in the context of MDS, that is, in the same process.
A SAN file system is logically composed of several entities on a SAN comprising clients, servers, and disks. SAN clients access file system object contents by accessing a pool of SAN disks used to store file system object contents. File system object metadata consists of naming information for pathname resolution and object attributes. One attribute associated with a file or directory, known as content-storage, describes how object contents are stored on a SAN disk via a storage object identifier (e.g., a list of block addresses). A content-storage attribute of an object allows a SAN client to directly read data from and write to SAN disks. Thus, a remote container module (RCM) treats requests for a content-storage attribute as requests for file contents. Storage for local copies is allocated and managed by a cache manager (CM). Newly allocated storage space is filled by a Distributed File Manager (DFM), wherein file system objects are copied from a remote source into a SAN disk.
A remote file system is inserted into local file system namespace using an attachment, where an attachment marks the border in the namespace between local data and local copies of remote data maintained by DST. Objects in namespace below an attachment associated with a remote source are said to reside in a remote container. DST intercepts metadata requests for remote container objects by utilizing previously mentioned modules while allowing MDS to continue managing local containers.
DST delegates fetching and storing operations to a Remote Access Agent (RAA), which facilitates direct data transfer between remote sources and SAN disks. An RAA is controlled by portions of a DST running on an MDS and handles protocol implementation and conversion necessary for communication with remote data sources. In accordance with a consistency level in a policy database, RAA fetches and updates local copies if a remote source indicates or RCM determines that a file system object has been modified since the time it was first stored on a SAN disk. DFM and RCM obtain necessary policy information for RAA via communications with a consistency policy manager.
Additionally, RAA returns to MDS metadata comprising file system object attributes, and directory mappings. Thus, RAA acts as a portal for MDS to a remote source. RCM creates an object locally corresponding to a requested remote file system object and sets its attributes to match those of requested remote file system object with exception of a content-storage attribute, which is adjusted to refer to locations in local storage. RAA fills local storage locations referred to by a content-storage attribute with file contents obtained from a remote source. A query to MDS allows a SAN client to obtain metadata corresponding to a requested file system object. A SAN client subsequently accesses a SAN disk, independently of any other module. In this manner, the transfer of file contents is abstracted from MDS.
While this invention is illustrated and described in a preferred embodiment, the invention may be produced in many different configurations. There is depicted in the drawings, and will herein be described in detail, a preferred embodiment of the invention, with the understanding that the present disclosure is to be considered as an exemplification of the principles of the invention and the associated functional specifications for its construction and is not intended to limit the invention to the embodiment illustrated. Those skilled in the art will envision many other possible variations within the scope of the present invention.
MDS 100 is further comprised of client interface 104, Replica Manager (RM) 106, RCM 108, MDS Code 110, CM 112, Distributed File Manager (DFM) 114, and Consistency Maintenance Framework (CMF) 116. Further comprising DST 124, outside of MDS 100, is RAA 120. SAN client 102 requests data through client interface 104 of MDS 100. Client interface 104 forwards remote container request to RCM 108. RCM 108 determines whether requested object is available as a local copy by checking metadata describing the storage of local copies in SAN disks, or a content-storage attribute. If a requested object is present as a local copy, its validity is verified in accordance with a consistency policy stored in CMF 116. A validity check performed by RCM 108 is comprised of two parts, a local check and an optional remote check. A local check determines either the existence of a consistency guarantee or that remote source 122 has been consulted within a specified time bound. Steps implemented in a local check are obtained from CMF 116. However, if a consistency guarantee does not exist or remote source 122 has not been consulted recently, communication with remote source 122 via DFM 114 and RAA 120 is initiated. Remote source 122 is contacted to determine whether a file system object has been modified at remote source 122, and if it has, a corresponding local copy of the object is marked as being invalid.
If the requested object is not valid or not present as a local copy, RCM 108 forwards an object request to DFM 114 which in turn forwards object data request to RAA 120. Thus, DST 124 obtains requested file system object from remote source 122. DST 124 component, RAA 120, fetches file system object contents and attributes from remote source 122. Furthermore, RAA 120 performs remote source 122 operations comprising: reading contents of a directory at remote source 122, reading object contents from a storage location on SAN disk 118, writing to a file system object on remote source 122, updating a directory on remote source 122 by creating or removing an entry, and reading and updating attributes associated with a file system object on remote source 122. Attributes associated with a fetched remote file system object stored in MDS 100 by RCM 108 and object content is stored directly in a SAN disk 118 at a location previously specified in a content-storage attribute in metadata obtained from RCM 108.
Remote file system objects from remote source 122 are held in a remote container, in which they are organized into a hierarchy with a root directory. A remote container is a local entity of MDS 100 managed by RCM 108. While all remote file system objects from remote source 122 are organized in a hierarchy and thus related to each other, it is not necessary for each to exist as a local copy. A remote file server has the capability to export multiple sub-trees and hence, be locally represented as multiple remote containers in a SAN file system. Cached file system objects and aggregates of related, remote file system objects, known as replicas, are stored in remote containers; a singular difference lies in the local population methods of remote containers holding cache copies versus those holding replicas. RM 106 manages consistency of file system objects that require asynchronous validation. CM 112 performs storage allocation and bookkeeping functions for RCM 108. File system object content is written to and stored in SAN disk 118 via RAA 120 without using MDS 100 as an intermediary.
Finally, to return a requested object to SAN client 102, DST 124 returns control to MDS 100, which allows SAN client 102 to access a local copy of requested data from SAN disk 118. Modifications to file system objects made locally are updated at remote source 122 in accordance with a policy specified in CMF 116. To preserve consistency of local copies of remote data with respect to changes made by external clients, SAN client 102 requests a consistency guarantee from MDS 100. MDS 100 requests a similar guarantee from DST 124, which in turn obtains a consistency guarantee from remote source 122. The protocol in which DST 124 communicates with remote source 122 determines the format and content of a consistency guarantee, for example, a time-based consistency guarantee.
The present invention arranges SAN file system contents in containers, specifically remote containers hold DST-managed data originating from remote sources. Each container comprises one or more directories and each directory further comprises one or more files, attachment points, and symbolic links. A directory contains subsidiary directories, each of which is identified by a name that is unique within its parent directory. The association between a directory name and a file system object is known as a mapping. A single directory within each container is noted as a root directory and is not a subsidiary of any directory within a container. A distinguished container provides a system root for a SAN file system. Other directories in a container are present in exactly one other directory in a given container. These uniqueness rules do not apply to “.” and “..”, current and parent directories, respectively. An attachment point connects a name in a directory of one container to the root directory of another container. A file system object therefore has a pathname, which is a list of directory names starting with a file system root and traversing directories and container attachments to the last name on a list identifying a file system object. Each file system object also has associated attributes. Of particular interest are file length and content-storage attributes; a block map or storage object identifier can be obtained from the latter. File system object attributes and directory mappings are collectively referred to as metadata and are stored in MDS 100.
Prior to direct operations on file system object contents, operations on metadata occur.
Collectively shown in
RCM 308 stores object and content-storage attributes, attributes for validating object contents, and directory mappings, collectively referred to as metadata, pertinent to file system objects stored in remote containers. Client interface 304 of MDS 300 passes to RCM 308 requests received from SAN client 320 for metadata for a file system object, as well as requests for a consistency guarantee or guarantees associated with a requested file system object. RCM 308 obtains a consistency guarantee pertaining to a requested file system object from DFM 314 and relays said consistency guarantee to client interface 304 of MDS 300. Thus, MDS 300 is able to extend equivalent guarantees to requesting SAN client 320. Client interface 304 of MDS 300 is also used by RCM 308 to deliver notification of changes in remote objects received from DFM 314 back to SAN client 320. In one embodiment, SAN client 320 is notified of changes by lock revocation.
CM 312 manages local storage for remote file system object contents. CM 312 supports both caching and replication functions. CM 312 implements a replacement algorithm identifying a subset of remote objects that are to be stored locally. RCM 308 communicates with CM 312 to request storage space for remote file system objects. CM 312 operates on portions of remote object content managed by RCM 308, which are either parts of files, whole files, or entire replicas, depending on a specified level of granularity. Because RCM 308 contains metadata describing the storage of local copies in SAN disks, and thus, determine local copy presence without consulting CM 312, CM 312 also communicates with RCM 308 to indicate the reclamation of storage space and subsequent eviction of stored remote file system objects. Additionally, CM 312 provides a mechanism for attaching to each chunk, pertinent subset of object attributes. Such a subset of object attributes is used by a replacement algorithm for determining the best subset to store. A cache replacement algorithm is influenced by policy input obtained from CMF 316. For example, priorities governing competition for cache space in CM 312 are adjusted in accordance with policy input obtained from CMF 316. Actual block allocation for a storage area governed by CM 312 is handled by MDS 300, specifically by server utilities in MDS Code 110, while an object-based storage system performs block allocation in SAN disk 322.
RM 306 operates in a manner different from that of client interface 104, managing replicas by scheduling the revalidation of their contents. Replicas are well-defined subsets of remote data, which have common policies for validation and update consistency. CMF 316 creates replicas as a result of defining policies on groups, or aggregates, of remote objects. Replica management involves the definition of replicas and the processing of asynchronous validations for local copies of remote file system objects.
DFM 314 communicates with RCM 308 to provide an interface to remote source 324. DFM 314 provides a method for accessing file system objects from remote source 324 and managing consistency guarantees encapsulating protocol-specific details. To implement such a remote data interface, DFM 314 communicates with RAA 326, which implements a protocol necessary to communicate with remote source 324. DFM 314 interface provides methods for operations on directories (e.g., lookup and create), obtaining and modifying file system object attributes, obtaining and returning consistency guarantees, and indirectly reading and writing object contents. While RCM 308 is not affected by protocol-specific details, each implementation of DFM 314 and RAA 326 address the concerns of a single protocol. In one embodiment, RCM 308 connects to a plurality of DFM 314 implementations further connecting to a plurality of RAA 326 implementations. Thus, communication through a plurality of protocols for diverse remote sources 324 is enabled.
CMF 316 records administrative preferences for important aspects of DST 318 behavior. Namely, preferences are consistency policies for validation and update transmission for local copies of remote objects. Validation and update parameters are supplied to DFM 314, while validation type (e.g., asynchronous or synchronous) is supplied to RM 306 and RCM 308. For each file system object in a remote container, a consistency policy is specified and associated. Consistency policies are maintained in CMF 316 along with entries binding consistency policies to file system objects. Consistency policies are independent and reusable objects, while policy entries are keyed by a remote container identifier and a relative pathname of a requested file system object within a keyed remote container. For efficiency of application and specification, consistency policies are defined using a hierarchy of aggregates comprising directories, sub-trees and remote containers. More detailed specifications supersede less detailed specifications that occur at a higher level in a file system hierarchy. Additionally, consistency policies are enabled based on file system object attributes, for example, name and size.
Network file access protocols vary in their provision of support for maintaining the consistency of remote data accessed by a client. To conceal support for varying levels of consistency maintenance from RCM 308, DFM 314 utilizes a mechanism that provides for consistency guarantees by using delegation and callbacks. A delegation corresponds to a consistency guarantee; it is a request to be notified of certain changes. A callback corresponds to such a notification. Delegations are designated for individual file system objects and for aggregates (e.g., directories and entire containers). In one embodiment, a delegation is typed and is applied to a specific operation, thus allowing separate control of read and write access to either metadata or object contents. In another embodiment, a delegation has multiple types and applies to a set of operations on either metadata or object contents. Delegation and callback mechanisms allow network file access protocols providing strong consistency guarantees to be supported without loss of function by mapping these consistency guarantees, while DFM 314 simulates the behavior for network file access protocols providing weaker consistency guarantees. The fidelity of such a simulation is controlled by consistency policy input from CMF 316. Network file access protocols also differ in support for other features, such as file system object attributes. DFM 314 also encapsulates these differences. The separation of functions between DFM 314 and RAA 326 is protocol and implementation dependent.
RAA 326 is invoked by DFM 314 to contact a requested remote data source 324. RAA 326 provides for reading file system object contents and attributes from remote source 324, returning of metadata to DFM 314, writing of file contents to SAN disk 322, and managing authentication and protocol state for DFM 314. A key feature of RAA 326 interface with DFM 314 is the abstraction of file contents from MDS 300. Rather than handling file contents, DFM 314 provides as input to RAA 326 a content-storage attribute (e.g., storage object identifier and offset, range of block addresses) indicating a location in SAN disk 322 in which requested file contents are stored. RAA 326 either writes received remote file contents to specified block addresses or reads from block addresses on SAN disk 322 locally modified file content. RAA 326 performs a read operation when locally modified content from specified block addresses is to be exported to remote source 324 from which it initially originated.
Consistency Maintenance Process Flow—Freshness Check and Fetching
The task of consistency management is shared between RCM 308, DFM 314, and CMF 316. Referring now to
If a consistency guarantee for a referenced local copy does not exist, attributes corresponding to referenced local copy of remote data are requested from DFM 314 in step 406. Subsequently, in step 408 remote object attributes are fetched from remote source 324. If an existing, revoked consistency guarantee for a referenced local copy is notification-based, a new consistency guarantee is obtained from remote source 324 in step 410. Otherwise, if it is time-based, a local timer is set in step 412. A time-based consistency guarantee is automatically revoked when the time interval set by a local timer has passed, whereas a notification-based consistency guarantee remains in effect until it is revoked upon receipt of an explicit notification from remote source 324. The process proceeds to step 414 where newly fetched attributes are merged with previous attributes. Merged attributes may cause the invalidation of directory or file content. For example, if a modification time attribute of a file has changed, associated data content is invalidated.
The process returns to step 402, where it is determined whether object contents of a referenced local copy have been requested; if such is the case, object content is validated in step 500. Otherwise, the process terminates with step 404.
Asynchronous and synchronous methods of validation are compatible with each other and thus, it is possible to specify both methods of validation in a single, hybrid policy. In one embodiment, asynchronous, synchronous, and hybrid policy types specified for different remote containers operate simultaneously within a single instantiation of MDS 300. For synchronous validation, RCM 308, connected to client interface 304, intercepts references to local copies of remote objects and performs validation before allowing requested data content or attributes to be returned to SAN client 320. When synchronous validation is not specified, MDS 300 is connected to a local copy through a null remote container module and asynchronous validation occurs as a background process, namely, as a background RM 306 process.
Consistency Maintenance Process Flow—Validation
Referring now to
Differences between remote directory structure and local containers, such as non-resident files, are rectified upon demand by RCM 308 such that objects represented in a container appears the same as SAN-resident data. If a referenced file system object is invalid or non-resident, referenced file system object content and metadata comprising directory mappings and attributes are obtained from remote source 324.
Consistency Maintenance Process Flow—Object Importation Process
In step 600, the type (e.g., file, directory) of an invalid object is checked. If it is determined that SAN client 320 has requested a content storage attribute (e.g., storage object identifier and offset, range of block addresses) for non-resident or invalid file contents, RCM 308 is invoked to request storage space from CM 312 in step 602. If sufficient storage space currently exists, step 602 is bypassed; this information is determined from metadata comprising file length and content-storage attributes. The process continues to step 604 where RAA 326, invoked by DFM 314, retrieves requested file data comprising file contents and associated attributes from remote source 324. RAA 326 stores retrieved file contents in SAN disk 322 in a storage location designated by RCM 308 in step 606. In step 608, RCM 308 writes the storage location in which file content was stored to a content-storage attribute in metadata corresponding to retrieved file data content. The process continues with step 610 where retrieved file contents are labeled with current file attributes, and thus, file content importation portion of the process is complete. In step 612 SAN client 320 obtains content-storage attribute from RCM 308 and accesses valid file content on SAN disk 322. In step 626, the process terminates.
Returning to type-checking step 600, if invalid object is of directory type, the process proceeds to step 614 where a requested directory is retrieved from remote source 324 via DFM 314. Remote source 324 returns a list of names and attributes for each child object in a directory sub-tree. In step 616, it is determined whether a local entry exists for a name selected from list returned in step 614. If such an entry does exist, the process proceeds to step 622; otherwise, the process proceeds to step 618 where a local object is created to represent a child object referred to by a name chosen in step 616. A local name is also associated with a newly created local object. Continuing to step 620, a local entry mapping local name to local object is created in a local directory for a new object. In step 622, which follows directly after step 616 if a local entry for already exists for a selected name, newly created local object is labeled with attributes associated with a child object it represents. The process returns to step 616 and repeats steps 616 through 622 for each name in list. After each name in list has been visited, directory content importation portion of the process is complete. The process continues to step 624 in which SAN client 320 accesses a requested, valid directory object in 624, and the process terminates in step 626.
A retrieved file system object is also accompanied by a consistency guarantee for a requested file system object, or a larger aggregate containing a consistency guarantee for a requested file system object. RCM 308 retains a consistency guarantee to ensure the continuity of file system object validity. Because file system object metadata is necessary to represent local storage details of file system object contents, it is necessary to purge file system object contents from local storage before corresponding metadata is purged. DFM 314 is used to request directory mappings and attributes for objects individually (e.g., by looking up a single pathname) or in bulk (e.g., by reading an entire directory). Results of a request from DFM 314 are converted into a form compatible with SAN file system metadata representation by RCM 308 of a data importation process.
Consistency Maintenance Process Flow—Object Exportation Process
Referring now to
If modifications to local copy marked in step 700 affect file contents, exportation process continues to step 702. In step 702, RAA 326 obtains metadata comprising a content storage attribute and file attributes corresponding to a modified local copy from RCM 308 via DFM 314. In step 704, RAA 326 retrieves file content stored in SAN disk 322 at location indicated by obtained content-storage attribute. RAA 326 exports file content and file attributes to remote source 324, as shown in step 706. A schedule determining the frequency and time interval at which modification updates are sent to a remote source is dependant on an update consistency policy setting. Update consistency policy configurations are defined by a delay (i.e. write-behind), which may be zero (i.e. write-through), or by using a method of delegation in which a remote source agrees to notify RCM 308 via RAA 326 and DFM 314, when remote source 324 requires an update. These considerations control the delay between the time at which a file system object is modified and when modification updates are sent back to a remote source. In one embodiment, delegation is used to provide single system image semantics. Continuing to step 708, remote source 324 returns remote file attributes and a consistency guarantee to RAA 326, which is subsequently returned as metadata for storage in MDS 300 by RAA 326 to RCM 308 via DFM 314. File content exportation process terminates in step 718.
If modifications to marked local copy in step 700 affect metadata, directory content exportation process continues to step 710 where RAA 326 obtains changed directory entries from RCM 308 via DFM 314 for marked local copies. In step 712, RAA extracts from directory entries and reformats, a name and a mapping corresponding to a local directory. Following in step 714, RAA 326 exports directory attributes and name changes obtained in step 712. Proceeding step 716, remote source 324 returns remote directory attributes and a consistency guarantee to RAA 326, which is returns directory attributes and a consistency guarantee as metadata for storage in MDS 300 by RAA 326 to RCM 308 via DFM 314. Directory exportation process terminates in step 718.
Remote Access Agent
RAA 326 handles protocol-specific details necessary to manage communications with a remote source. RAA 326 implements network file access protocols used to communicate with remote source 324 and directs transfer of file content between remote sources and SAN disks. A key feature of RAA 326 interface with DFM 314 is that it allows RCM 308 to avoid handling file contents. Instead, DFM 314 passes a content-storage attribute (e.g., storage object identifier and offset, range of black address) obtained from CM 312 to RAA 326. RAA 326 uses block addresses to write to SAN disk 322 contents of remote file system objects it receives, as well as to read locally modified data from SAN disk 322 and send it to remote source 324.
RAA 326 code is linked with library code that allows data transfer to and from remote source 324 using a variety of network file access protocols. In one embodiment, RAA 326 implementation is specialized for given network file access protocols and specific remote sources. Data transfer is accomplished by making direct calls to network file access client code and by making file input/output (I/O) system calls to a remote operating system kernel. RAA 326 is indistinguishable from a typical network file access client from the perspective of remote source 324. RAA 326 also utilizes consistency mechanisms supported by network file access protocol and maintains these mechanisms in proxy for its caller, namely DFM 314. RAA 326 also acts as a security proxy for SAN client 320. Depending on the security protocol in use, RAA 326 receives credentials for a particular user and uses them as necessary to perform authentication. In this manner, authorization to perform a particular file operation is obtained.
In one embodiment, several agents are instantiated, allowing for an improvement in throughput and parallelism. In another embodiment, an agent is implemented with internal parallelism; it is comprised of sub-agents implemented on a plurality of distinct machines. The separation of the DFM 314 and RAA 326 necessitates inter-process communication between machines hosting each module. Because DFM 314 interface uses function calls and callbacks, an RPC connection or connections allowing bi-directional calls are necessary.
Additionally, RAA 326 makes provisions for reading and writing metadata and file contents, as well as managing authentication and protocol state for DFM 314. RAA 326 acts as a data-handling proxy serving as an intermediary between MDS 300, remote source 324, and SAN disk 322. A protocol-specific implementation of RAA 326 is invoked by DFM 314 to initiate communications with remote source 324. DFM 314 selects an implementation of RAA 326 based on source remote container and RAA 326 configuration. DFM 314 sends requests to RAA 326; RAA 326 receives requests and identifies appropriate remote host, protocol, and user authentication for communication. In one embodiment, RAA 326 responds immediately by providing a file system object handle for a known path. Otherwise, RAA 326 sends request to remote source 324 and processes returned response. Depending on network file access protocol, local protocol state is also managed by RAA 326. Local protocol state processing is comprised of maintenance of: file handle and lock owner identifiers, sequence numbers for serializing operations, and callbacks for consistency management.
Additionally, the present invention provides for an article of manufacture comprising computer readable program code contained within implementing one or more modules to locally store, share, and maintain remote file system data. Furthermore, the present invention includes a computer program code-based product, which is a storage medium having program code stored therein which can be used to instruct a computer to perform any of the methods associated with the present invention. The computer storage medium includes any of, but is not limited to, the following: CD-ROM, DVD, magnetic tape, optical disc, hard drive, floppy disk, ferroelectric memory, flash memory, ferromagnetic memory, optical storage, charge coupled devices, magnetic or optical cards, smart cards, EEPROM, EPROM, RAM, ROM, DRAM, SRAM, SDRAM, or any other appropriate static or dynamic memory or data storage devices.
Implemented in computer program code based products are software modules for: (a) storing structured local copies of remote data; (b) storing metadata pertaining to said remote data; (c) validating local copies; (d) accessing local copies utilizing said metadata; (e) importing data from a remote source when changes to remote data occur; and (f) exporting data to a remote source when changes to local copies occur.
A system and method has been shown in the above embodiments for the effective implementation of a remote access agent for caching in a SAN file system. While various preferred embodiments have been shown and described, it will be understood that there is no intent to limit the invention by such disclosure, but rather, it is intended to cover all modifications falling within the spirit and scope of the invention, as defined in the appended claims. For example, the present invention should not be limited by software/program or specific computing hardware.
The above enhancements are implemented in various computing environments. For example, the present invention may be implemented on a multi-nodal system (e.g., LAN, SAN, WAN, MAN) or networking system (e.g., Internet, WWW). All programming and data related thereto are stored in computer memory, static or dynamic, and may be retrieved by the user in any of: conventional computer storage and display (i.e., CRT). The programming of the present invention may be implemented by one of skill in the art of object-oriented and network programming.